This book is unique in that it contains research material from two, until now, relatively distinct and separate communities: statisticians working on robust techniques for model fitting and for model selection and, as a largely separate community, computer scientists and engineers working on problems involving data segmentation. In the last few years there has been a growing awareness, by some members of both communities, that there are techniques of common interest. Engineers and computer scientists increasingly turned to statistical techniques for guidance in constructing algorithms for solving real world problems such as segmentation of range image data (to locate and or identify objects of interest in a scene) and for segmentation of image sequences based upon motion. However, they found that these problem domains demanded modified or extended versions of the techniques used by statisticians. Likewise, some statisticians have recently discovered that the problems associated with machine vision tasks provided new challenges and began to look at such areas as a source of new inspiration.

Until this monograph, the research results of the research of each community tended to appear only in the separate publishing avenues traditionally used by each community. Thus, apart from some very recent special issues of journals, there has been no common publications where the work from both communities are collected together and made readily accessible to both communities. The purpose of this monograph, therefore, is to help fill that gap.

To the statistician, not only will this monograph contain a convenient summary of recent work in their community, but it will provide them with an excellent primer on the problems of data segmentation within a machine vision context. This will help them to quickly learn what has been done, in this area, what are the important aspects of the problems, and what are some of the potential applications of techniques related to their own expertise. It is hoped the monograph will inspire them and help them to work in these areas.

To the engineer and computer scientist, this monograph will provide a convenient entry point into current research in data segmentation as well as providing a synopsis of the main statistical ideas relevant to a study of this subject within a statistical context.

The material is presented at a level that is appropriate for senior undergraduates, doctoral and masters thesis students, and established researchers in either community. It may serve as a resource for a seminar series at postgraduate level.

Each chapter of the book is written by a leading authority in the relevant areas. Chapter 1 provides a historical overview of data segmentation in machine vision, and is written by Prof. R. Jarvis of the Intelligent Robotics Research Center, Monash University. Chapters 2-4 cover the statistical foundations: Chapter 2 is a summary of the robust statistical concepts by Prof. E. Ronchetti, Dept. of Econometrics, University of Geneva. Chapter 3 covers recent work by Dr S. Sommer and Assoc. Prof. R. Staudte , of the Dept. of Statistics at La Trobe University, in robust measures and model selection. Chapter 4 provides a tutorial on Geometric Inference and Model Fitting: a theory being specially developed by Prof. K. Kanatani, of Gunma University, in order to bridge the gap between the theories of statistics and demands of computer vision tasks. Finally, chapters 5 and 6 present work that uses statistical techniques for data segmentation. In chapter 5, Drs. A. Bab-Hadiashar and D. Suter of the Intelligent Robotics Research Center, Monash University, demonstrate techniques for range data segmentation and for optic flow segmentation. Chapter 6 contains the work of Dr. P. Torr of Microsoft Research, Cambridge UK, in model selection and segmentation of displacements between image sequences.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Springer-Verlag New York.